Intelligent identification method for main control factors of deformation and instability of surface mine slope

Abstract

The invention relates to an intelligent identification method for main control factors of surface mine slope deformation instability, and the method comprises the following steps: S1, data acquisition and preprocessing: obtaining surface mine slope geological data, deformation data and environmental data, and removing and filling abnormal values and missing values; s2, determining candidate control factors: extracting the candidate control factors from the maximum vibration rate in the environment data and the rainfall data; s3, key control factor identification based on a neighborhood rough set: calculating the importance degree of each control factor to the mine slope deformation data based on a neighborhood rough set method, and identifying the key control factors; and S4, key control factor verification based on support vector regression: establishing an SVR model of the screened key control factors and deformation data, and evaluating the accuracy of the key factors based on an SVR prediction effect. According to the method, the SVR model is established based on the neighborhood rough set and the support vector machine regression method, and the accuracy of the key factors can be evaluated based on the SVR prediction effect.

Description

Technical Field

[0001] This invention relates to the field of deformation evolution mechanism of open-pit mine slopes, and in particular to an intelligent identification method for the main controlling factors of deformation and instability of open-pit mine slopes. Background Technology

[0002] Open-pit mine landslides, as a type of disaster, are characterized by their suddenness, destructive power, and wide impact, seriously threatening people's lives and property and the ecological environment. The deformation evolution of open-pit mine landslides is a complex nonlinear process influenced by multiple factors. Deformation is a crucial criterion for judging the instability of open-pit mine slopes, and rainfall and blasting vibrations are important factors affecting open-pit mine instability. Therefore, accurately identifying the key control factors of landslide deformation is essential for the effective prevention and control of open-pit mine landslide disasters.

[0003] Traditional methods for identifying key factors in open-pit mine landslides mainly rely on expert experience and statistical analysis. Methods based on expert experience depend on knowledge of landslides in specific areas and are difficult to generalize to other regions. Existing model methods also have low accuracy in predicting key factors. Summary of the Invention

[0004] The purpose of this invention is to propose an intelligent identification method for the main controlling factors of slope deformation and instability in open-pit mines, thereby solving the problem of low accuracy of key factors in the evaluation of the prediction effect of traditional prediction methods.

[0005] This invention provides an intelligent identification method for the main controlling factors of slope deformation and instability in open-pit mines, comprising the following steps:

[0006] S1. Data Acquisition and Preprocessing: Acquire geological data, deformation data, and environmental data of open-pit mine slopes, and remove and fill out outliers and missing values.

[0007] S2. Candidate control factor determination: Extract candidate control factors from the maximum vibration rate and rainfall data in the environmental data;

[0008] S3. Identification of key control factors based on neighborhood rough set: The importance of each control factor to the mine slope deformation data is calculated based on the neighborhood rough set method to identify key control factors;

[0009] S4. Validation of key control factors based on support vector regression (SVR): Establish an SVR model of the selected key control factors and deformed data, and evaluate the accuracy of the key factors based on the prediction effect of SVR.

[0010] Preferably, in step S3, the decision system can be represented as {U, C∪D}, where U = {x1, x2, ..., x...} n} is a non-empty finite sample set, referring to all the collected data; C = {a1, a2, ..., a...} n} is the conditional attribute set; D is the decision attribute set, indicating whether it significantly causes slope deformation; considering the differences in order of magnitude and dimension of the data, normalization methods are used to preprocess the data to eliminate the influence of these differences; let x1 and x2 be N-dimensional spaces A = {x1, x2, ..., x...} N The two objects in}, f(x) i ,a j ) is sample x i a j Attributes, where i = 1, 2, ..., N; j = 1, 2, ..., N; the function Δ can usually be expressed as the Minkowsky distance:

[0011]

[0012] This formula is used to calculate the "difference" or "distance" between two data points, and can measure how much the "overall difference" is between any two data samples after considering all attributes;

[0013] P has three possible values ​​(1, 2, and ∞). When p = 1, the algorithm is to directly add the absolute values ​​of the differences of all attributes. When p = 2, the algorithm is to square the differences of all attributes, add them together, and then take the square root. When p = ∞, the algorithm is to take the one with the largest difference among all attributes as the total difference.

[0014] Considering x i ∈U and Neighborhood δ in feature space B B (x i This can be represented as:

[0015] δ B (x i )={x j |x j ∈U,Δ B (x i ,x j )≤δ} (2)

[0016] In the formula, δ is the neighborhood radius and δ(a i ) = std(a i ) / λ; λ is a constant value that varies according to the neighborhood precision, 2≤λ≤4;

[0017] Domain approximation space<U,N> For any X∈U, the upper and lower approximations of object X can be expressed as:

[0018]

[0019] Similar to traditional rough set theory, positive field POS B (D) and the lower approximation of the decision (POS) B (D)= N B D), that is, the neighborhood subset always belongs to a decision class;

[0020] Given a neighborhood decision system<U,C∪D,N> ,like And if a∈B, then the importance of conditional attribute a to decision attribute D relative to conditional attribute set B is:

[0021] Sig(a,B,D)=γ B (D)-γ B-a (D), (4)

[0022]

[0023] In the formula γ B (D) represents the dependence of decision attribute D on condition attribute B; |·| represents the cardinality of the set, i.e. the number of elements in the set; therefore, if Sig(a,B,D)≤ε, it means that attribute a in the condition attribute set is redundant for decision attribute D, otherwise a is indispensable; ε represents a small positive real number to control convergence; based on the above neighborhood rough set theory, the reduced set (red) without redundant attributes can be obtained by using the relevant attribute reduction algorithm.

[0024] Preferably, in step S4, Support Vector Machine Regression (SVR) is an extension of Support Vector Machine (SVM) in regression problems. Its goal is to find a regression function f(x) such that the deviation between the predicted value and the true value does not exceed a preset tolerance error ε, while ensuring the flatness of the model. The decision function expression of the SVR regression model is:

[0025]

[0026] In the formula W T The hyperplane coefficient vector; Let b be a nonlinear mapping function; b is the bias; the solution for minimizing the coefficients and bias term is expressed as:

[0027]

[0028] In the formula, D(f) is the optimal linear regression plane; ||W|| 2 R represents model complexity; C is the penalty factor; ε Let ε be the control error function; the optimization problem can be simplified to:

[0029]

[0030] In the formula ξj , It is a relaxation factor;

[0031] Using the Lagrange equation and duality theory, equation (8) is transformed into:

[0032]

[0033] In the formula K(x) r ,x j ) is the kernel function of the SVR model; a r ,a r * For Lagrange multipliers;

[0034] The optimal solution can be expressed as:

[0035] Using the key control factors identified by the neighborhood rough set as input variables and the daily displacement rate as output variable, an SVR prediction model was established to verify the accuracy of the identified control factors.

[0036] This invention discloses an intelligent identification method for the main controlling factors of open-pit mine slope deformation and instability. Deformation is an important criterion for judging the instability of open-pit mine slopes, and rainfall and blasting vibration are important factors affecting the instability of open-pit mines. Based on the neighborhood rough set and support vector machine regression (SVR) method, the maximum rainfall, cumulative rainfall, and maximum vibration rate in the x, y, and z directions of the open-pit mine slope in different time intervals are selected as candidate controlling factors. The method calculates the importance of different candidate controlling factors to the daily displacement of the slope based on the neighborhood rough set method to identify key controlling factors. An SVR model of the selected key controlling factors and deformation data is established, and the accuracy of the key factors is evaluated based on the SVR prediction effect. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the calculation process of the intelligent identification method for the main controlling factors of slope deformation and instability in open-pit mines, as described in this invention.

[0038] Figure 2 This is a map of mine slope monitoring data according to an embodiment of the present invention.

[0039] Figure 3 This is a verification diagram of the deformation characteristic factors of mine slope based on SVR according to an embodiment of the present invention.

[0040] in Figure 3 In the middle, (a) H01; (b) H02; (c) H03. Specific implementation methods

[0041] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings. Before formally describing the present invention, a general description of the solution of the present invention will be given first to facilitate understanding.

[0042] An embodiment of an intelligent identification method for the main controlling factors of slope deformation and instability in open-pit mines includes the following steps:

[0043] S1. Data Acquisition and Preprocessing: Acquire geological data, deformation data, and environmental data of the open-pit mine slope; remove and impute outliers and missing values. As one example, removing missing values ​​specifically means directly deleting data with a missing value percentage of less than 5%. Handling outliers specifically means removing existing outliers from numerical outlier data and interpolating to impute missing values ​​before and after them.

[0044] S2. Candidate control factor determination: Extract candidate control factors from the maximum vibration rate and rainfall data in the environmental data;

[0045] S3. Identification of key control factors based on neighborhood rough set: The importance of each control factor to the mine slope deformation data is calculated based on the neighborhood rough set method to identify key control factors;

[0046] S4. Validation of key control factors based on support vector regression (SVR): Establish an SVR model of the selected key control factors and deformed data, and evaluate the accuracy of the key factors based on the prediction effect of SVR.

[0047] Attribute reduction is used to obtain a sub-attribute set (or relative reduction) with the same discriminative power as the complete attribute set. Pawlak first proposed rough set theory, which has since been widely applied to feature selection and other fields. Similar to common methods for identifying the main controlling factors of slope deformation, such as grey relational analysis and decision trees, classical rough set theory is only applicable to categorical data. Numerical data, such as monitoring data, requires data discretization using methods like cluster analysis. However, most real-world data is numerical, and some important information may be lost during discretization. Furthermore, different discretization methods can significantly affect the analysis results. Neighborhood rough set theory can be used in both attribute reduction and classification to handle both numerical and categorical data.

[0048] In step S3, the decision system can be represented as {U, C∪D}, where U = {x1, x2, ..., x}. n} is a non-empty finite sample set, referring to all the collected data; C = {a1, a2, ..., a...} n} is the conditional attribute set, as shown in Table 1; D is the decision attribute set, representing whether it significantly causes slope deformation; considering the differences in order of magnitude and dimension of the data, normalization methods are used to preprocess the data to eliminate the influence of these differences; let x1 and x2 be N-dimensional spaces A = {x1, x2, ..., x...} N The two objects in}, f(x) i ,a j ) is sample x i a j Attributes, where i = 1, 2, ..., N; j = 1, 2, ..., N; the function Δ can usually be expressed as the Minkowsky distance:

[0049]

[0050] This formula is used to calculate the "difference" or "distance" between two data points, and can measure how much "overall difference" exists between any two data samples after considering all attributes (rainfall, vibration, etc.).

[0051] P has three possible values ​​(1, 2, and ∞). When p = 1 (Manhattan distance), the algorithm is to directly add the absolute values ​​of the differences of all attributes. When P = 2 (Euclidean distance), the algorithm is to square the differences of all attributes, add them together, and then take the square root. When P = ∞ (Chebyshev distance), the algorithm is to take the one with the largest difference among all attributes as the total difference.

[0052] Considering x i ∈U and Neighborhood δ in feature space B B (x i This can be represented as:

[0053] δ B (x i )={x j |x j ∈U,Δ B (x i ,x j )≤δ} (2)

[0054] In the formula, δ is the neighborhood radius and δ(a i ) = std(a i ) / λ; λ is a constant value that varies according to the neighborhood precision, 2≤λ≤4; in this embodiment, it is taken as 2.

[0055] Given a neighborhood approximation space<U,N> For any X∈U, the upper and lower approximations of object X can be expressed as:

[0056] Similar to traditional rough set theory, positive field POS B(D) and the lower approximation of the decision (POS) B (D)= N B D), that is, the neighborhood subset always belongs to a decision class;

[0057] Given a neighborhood decision system<U,C∪D,N> ,like And if a∈B, then the importance of conditional attribute a to decision attribute D relative to conditional attribute set B is:

[0058] Sig(a,B,D)=γ B (D)-γ B-a (D), (4)

[0059]

[0060] In the formula γ B (D) represents the dependence of decision attribute D on condition attribute B; |·| represents the cardinality of the set, i.e. the number of elements in the set; therefore, if Sig(a,B,D)≤ε, it means that attribute a in the condition attribute set is redundant for decision attribute D, otherwise a is indispensable; ε represents a small positive real number to control convergence, which is taken as 0.1 in this embodiment; based on the above neighborhood rough set theory, the reduced set (red) without redundant attributes can be obtained by using the relevant attribute reduction algorithm.

[0061] Among them, such as Figure 1 As shown in step S3, the relevant attribute reduction algorithm is as follows:

[0062]

[0063] In step S4, Support Vector Machine Regression (SVR) is an extension of Support Vector Machine (SVM) in regression problems. Its goal is to find a regression function f(x) such that the deviation between the predicted and true values ​​does not exceed a preset tolerance error ε, while ensuring the flatness of the model. The decision function expression of the SVR regression model is:

[0064]

[0065] In the formula W T The hyperplane coefficient vector; Let b be a nonlinear mapping function; b is the bias; the solution for minimizing the coefficients and bias term is expressed as:

[0066]

[0067] In the formula, D(f) is the optimal linear regression plane; ||W|| 2 R represents model complexity; C is the penalty factor; εLet ε be the control error function; the optimization problem can be simplified to:

[0068]

[0069] In the formula ξ j , It is a relaxation factor;

[0070] Using the Lagrange equation and duality theory, equation (8) is transformed into:

[0071]

[0072] In the formula K(x) r ,x j ) is the kernel function of the SVR model; a r ,a r * For Lagrange multipliers;

[0073] The optimal solution can be expressed as:

[0074] Using the key control factors identified by the neighborhood rough set as input variables and the daily displacement rate as output variable, an SVR prediction model was established to verify the accuracy of the identified control factors.

[0075] Application examples:

[0076] Figure 2 The data from on-site monitoring of a mine slope shows that surface displacement at each monitoring point exhibited a continuous increasing trend under the combined effects of blasting vibration and rainfall. From May 31 to June 3, 2025, the mine slope experienced continuous heavy rainfall, and the maximum daily vibration rate increased sharply, causing the surface displacement to enter a rapid deformation phase. By the end of the monitoring period, the cumulative displacements at each monitoring point were 319.96 mm, 333.52 mm, and 286.07 mm, respectively.

[0077] The main controlling factors of mine slope deformation characteristics were identified based on 16 candidate control factors. As shown in Table 1, these candidate control factors are rainfall and maximum vibration rate conditions at different time intervals. The importance of each control factor to the surface displacement at different monitoring points of the mine slope is shown in Table 2.

[0078] Table 1 Candidate Control Factors for Mine Slope Deformation Monitoring Data

[0079]

[0080] Table 2. Importance of Conditional Attributes to Mine Slope Monitoring Data

[0081]

[0082] Table 2 Importance of Conditional Attributes to Mine Slope Monitoring Data (Continued)

[0083]

[0084] Note: Space indicates that the control factor has an importance of less than 0.1 to the monitoring data; bold values ​​indicate the control factor with the most significant impact on the monitoring data.

[0085] The validation dataset shows that, Figure 3 As shown, the SVR prediction effect is good, and the prediction model R between deformation data at different monitoring points and key control factors is satisfactory. 2 The value is close to 0.9, indicating that the key control factors screened based on neighborhood rough set have a certain degree of effectiveness.

Claims

1. An intelligent identification method for main control factors of slope deformation and instability in an open-pit mine, characterized in that The method comprises the following steps: S1, data acquisition and preprocessing: obtaining open-pit mine slope geological data, deformation data and environmental data, removing and filling abnormal values and missing values; S2, candidate control factor determination: extracting candidate control factors from the maximum vibration rate in the environmental data and rainfall data; S3, key control factor identification based on neighborhood rough set: calculating the importance of each control factor on the mine slope deformation data based on the neighborhood rough set method, and identifying the key control factor; S4, verification of key control factors based on support vector regression (SVR): establishing an SVR model of the screened key control factors and deformation data, and evaluating the accuracy of the key factors based on the SVR prediction effect.

2. The intelligent identification method of the main control factor of open-pit mine slope deformation instability according to claim 1, characterized in that: In step S3, the decision system can be represented as {U, C∪D}, where U = {x1, x2, …, x n} is a non-empty finite sample set, which refers to all the collected data; C = {a1, a2, …, a n} is a set of condition attributes; D is a set of decision attributes, which is whether to significantly cause slope deformation; Considering the differences in the order of magnitude and dimension of the data, the data is preprocessed by using the normalization method to eliminate the influence of these differences; Let x1 and x2 be N-dimensional spaces A = {x1, x2, ..., x} N The two objects in}, f(x) i ,a j ) is sample x i a j Attributes, where i = 1, 2, ..., N; j = 1, 2, ..., N; the function Δ can usually be expressed as the Minkowsky distance: The formula is used to calculate the "difference degree" or "distance" of two data points, and can measure the "comprehensive difference" between any two data samples after considering all attributes; P has three numerical values (1, 2 and ∞), when p = 1, the algorithm: directly add the absolute value of the difference of all attributes; when P = 2, the algorithm: square all attribute differences and then take the square root; when P = ∞, the algorithm: only take the largest difference in all attributes as the total difference; Considering x i ∈ U and The neighborhood δ B (x i ) in the feature space B can be represented as: delta B (x i ) = {x j |x j ∈ U, delta B (x i , x j ) ≤ delta} (2) where δ is the neighborhood radius and δ(a i ) = std(a i ) / λ; λ is a constant value that varies according to the neighborhood accuracy, 2 ≤ λ ≤ 4; Given a neighborhood approximation space <U, N>, for any X ∈ U, the upper and lower approximations of object X can be expressed as: Similar to the traditional rough set theory, the positive region POS B (D) and the lower approximation of the decision (POS B (D) = N B D), i.e., where the neighborhood subset always belongs to one decision class; Given a neighborhood decision system <U, C∪D, N>, if and a∈B, the importance of the condition attribute a to the decision attribute D with respect to the condition attribute set B is: Sig(a, B, D) = γ B (D) - γ B-a (D), (4) where γ B (D) represents the dependency degree of decision attribute D on condition attribute B; | · | represents the cardinality of the set, i.e. the number of elements in the set; therefore, if Sig(a, B, D) < ε, it means that the attribute a in the condition attribute set is redundant for the decision attribute G, otherwise a is indispensable; ε represents a small positive real number to control convergence; based on the above neighborhood rough set theory, the reduction set (red) without redundant attributes can be obtained by using the relevant attribute reduction algorithm.

3. The intelligent identification method of the main control factor of open-pit mine slope deformation instability according to claim 2, characterized in that: In step S4, support vector machine regression (SVR) is an extension of support vector machine (SVM) in regression problems, and the goal is to find a regression function f(x) so that the deviation between the predicted value and the true value does not exceed the preset tolerance error ε, while ensuring the flatness of the model; The decision function expression of the SVR regression model is: where W T is a hyperplane coefficient vector; is a nonlinear mapping function; b is a bias term; solving for the coefficients and bias term minimizes the expression where D(f) is the optimal linear regression plane; ||W|| is the norm of the weight vector; and 2 is the model complexity; C is the penalty factor; R ε is the ε control error function; the optimization problem can be simplified as: wherein ξ j , is a relaxation factor; The Lagrange equation and the dual theory are used to convert formula (8) to: where K(x r ,x j ) is the SVR model kernel function; a r ,a r * are Lagrange multipliers; The optimal solution can be expressed as: The identified key control factors by neighborhood rough set are taken as input variables, and the daily displacement rate is taken as output variable to establish an SVR prediction model to verify the accuracy of the identified control factors.

Images